MySurgeryRisk Model Predictions of Postoperative Complications and Mortality

Area of Science:

• Medical Informatics
• Surgical Risk Prediction
• Health Services Research

Background:

• Postoperative complications affect up to 15% of surgical patients, underscoring the need for robust risk estimation models.
• Electronic health record (EHR) data across diverse settings can enhance surgical risk prediction accuracy.

Purpose of the Study:

• To test if the MySurgeryRisk framework, initially validated on a single center, maintains predictive performance and generalizability on a large multicenter dataset.
• To evaluate the framework's ability to predict intensive care unit (ICU) admission, mechanical ventilation (MV), acute kidney injury (AKI), and in-hospital mortality.

Main Methods:

• A retrospective, longitudinal, multicenter cohort study included over 500,000 patient encounters from 14 healthcare institutions.
• eXtreme Gradient Boosting models were developed using EHR data and validated for predicting key postoperative complications.
• Model performance was assessed using area under the receiver operating characteristics curve (AUROC).

Main Results:

• The study included 366,875 adult patients undergoing major inpatient operations.
• Prevalence of complications: 8% for ICU admission, 4% for MV, 7% for AKI, and 1% for in-hospital mortality.
• High AUROC values were achieved for all outcomes (0.92-0.95), comparable to the original MySurgeryRisk model. Procedure codes and clinician factors were the most influential variables.

Conclusions:

• A validated model using routinely collected EHR data accurately predicts postoperative complications and mortality across a large healthcare network.
• The MySurgeryRisk framework demonstrates generalizability, confirming its utility in diverse clinical settings.
• Procedure type and clinician-specific factors significantly influence surgical outcomes and risk prediction.